Lecture 2 Flashcards
What are the different types of psychological research?
- basic (look for knowledge for its own sake; theory) vs. applied (application)
- lab (control over research) vs. field (realism, face validity, ecological validity)
- quantitative (stats) vs. qualitative (richness of description)
What are the different experimental designs?
- single factor (2 levels, or more)
- factorial (2 or more, main effects and interactions)
- correlational (association)
- regression (predict)
What is the difference between applied and quasi-experimental research? What is the issue with this?
- applied is controlled, forced random assignment
- quasi-experimental is when groups occur naturally, no random assignment possible
ISSUE: lose any sense of causality. You have lost control
What are the problems with NHST? What should you do to attempt to remedy this?
- low power: may be a diff you don’t see
- high power: find even a very small diff as sig
- Type 1 error: false +ve; accepting false hypothesis as true
- Type 2 error: false -ve; rejecting true hypothesis as incorrect
- FIX: report CIs and effect sizes with p values
What are the two types of multivariate models we fit to data?
- confirmatory: seek to test prediction
- exploratory: models that seek to account for rships
What 3 things do you use to assess the fit of a model?
- residuals
- summary measures (eg. R2)
- statistical test (eg. null)
What are the different types of variables?
- independent vs. dependent (outcome vs. predictor)
- discrete vs. continuous (discrete can be categorical or orindal)
- exogenous vs. endogenous (exo = outside system we’re modelling)
What are the levels of measurement?
- nominal (categorical)
- ordinal
- interval
- ratio
What are the 2 types of missing data?
- missing completely at random (MCAR)
- missing at random (MAR)
- both of these are IGNORABLE
What is MCAR?
- missingness not related to any other variable
- BEST missingness
- Little’s test (want >.05), don’t reject null that it’s MCAR
What is MAR?
- missingness related to another variable, but no patterns within the variable (not insidious)
- prob. that missingness is unrelated, after controlling for another variable
- eg. depressed people less likely to report SES, but within depressed people the prob of reported SES is unrelated to SES level
How should you estimate missing data?
- DO NOT use the mean
- SPSS Missing Value Analysis: uses regression and EM algorithm approaches
- SPSS Multiple Imputation. MI creates multiple complete sets of data
- if a person is missing some items in a scale but has responded to others, then use the data you have to estimate missing data
When do you transform data?
- only if you really have to
- can transform so that it meets the assumptions of your test
- can change symmetry of data (i.e. want normal distro.) > can fix skew but not kurtosis
Why use Guttman instead of Alpha?
- alpha not the best
- all relaibility coefficients are pessimistic (lower bounds - i.e. that value or higher)
- Guttman allows for choice of lower bound estimates
How do we develop research?
- observations
- theory
- past research (gaps)
What are the essential elements of experimental research?
- hold some variables constant
- vary others (IV)
- observe changes on another variable (DV)
What are within and between subjects designs? What are the main issues with each of these?
- between: independent groups; need equivalent groups (use random sampling and random allocation)
- within: repeated measures; need to control for sequencing (learning) effects (counterbalance)
What is the goal of applied research?
to investigate real-world phenomenon
What 2 things should you never say?
- X causes Y
- X leads to Y
What is multivariate?
multiple DVs
What is the basic model that underlies everything?
data = model + residual
- have a residual for each individual or each cell
What does the residual account for?
- random noise
- measurement error
- diffs in scales
- etc.
- tells us how well the model FITS the data
What are the 2 types of NHST in model fit?
- model IS null: want less than .05, reject model/null, residuals large
- model NOT full: want >.05, do not reject model model, residuals not too large
What are the model parameters in the regression equation?
- the regression coefficients/weights
How can you represent the regression equation in matrix form? What does regression do?
Y = Xb + e
- predicting a vector from a matrix
What is the regression equation?
Y = b0 + b1X1 + b2X2 + e
What does i indicate in the regression equation?
- the model holds for every i
- holds for every person
What 4 things do you need to do after you imput your data?
- CHECK the data
- missing data
- transformations
- form totals
What are the 2 types of deleting data? Which one is better?
- pairwise: remove pairs of cases relevant to analysis
- listwise: remove any case with missing data (LOSE POWER!)
- pairwise is better, listwise loses power
What are the steps involved in missing data?
- determine extent
- delete cases
- evaluate missing values
What are common data transformations?
PULL IN HIGHER TAIL
- (-)1/x2 > strongest change
- (-)1/x
- log(x)
- sqrt(x)
PULL IN LOWER TAIL
- x2
- x3
- antilog(x) > strongest change
When is correlational/regressional research helpful?
- when experimental cannot be done
- cannot randomly assign people to groups
- has ecological validity
When is causality possible?
- carefully designed longitudinal studies
- careful experimental design
What type of variables do ANOVA and independent t-tests require?
discrete
What is interesting about ordinal scales in psychological research?
they are often assume to be interval
- this is usually fine, but can sometimes cause issues
What do totals involve? (6 things)
- reliability of test scores
- validity of subscales
- develop new subscales, if necessary
- form empirically weighted totals
- you can drop ‘dud’ items
- cope with missing data
